Research map / Video / Video segmentation and tracking
Read and Attend: Temporal Localisation in Sign Language Videos
Gul Varol, Liliane Momeni, Samuel Albanie, Triantafyllos Afouras, Andrew Zisserman
Check your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add attention-mined annotations to dictionary annotations | BSL-1K sign recognition · BSL-1K testrecnew | top-1 per-instance accuracy: 62.1 -> 64.2 | Improved |
| Add attention-mined annotations to mouthing annotations | BSL-1K sign recognition · BSL-1K testrecnew | top-1 per-instance accuracy: 40.8 -> 58.5 | Improved |
| Combine mouthing, dictionary, and attention annotations | BSL-1K sign recognition · BSL-1K testrecnew | top-1 per-instance accuracy: 63.5 -> 65.0 | Improved |
| Use greedy decoding with subtitle filtering instead of teacher forcing | Attention-mined recognition annotations · BSL-1K testrecnew | top-1 per-instance accuracy: 38.7 -> 53.9 | Improved |
| Remove subtitle filtering from greedy-decoded predictions (ablation) | Attention-mined recognition annotations · BSL-1K testrecnew | top-1 per-instance accuracy: 53.9 -> 50.6 | Hurt |
| Remove the requirement that training subtitles contain a sparse annotation (ablation) | Video-to-text Transformer · BSL-1K testloc | recall: 16.5% vs 6.8% | Hurt |
| Add sparse-annotation alignment loss (ablation) | Video-to-text Transformer · BSL-1K testloc | localisation accuracy (teacher forcing): 51.1% -> 58.7% | Improved |
| Add sparse-annotation alignment loss (ablation) | Video-to-text Transformer · BSL-1K testloc | recognition top-1 per-instance accuracy: 53.9 vs 53.6 | No clear effect |
| Add curriculum from one-sign trimmed clips to subtitle-length inputs (ablation) | Video-to-text Transformer · BSL-1K testloc | recall: 15.8 -> 16.0 | Improved |
| Use active-signer detection and sparse annotations to realign subtitles (ablation) | Video-to-text Transformer · BSL-1K testloc | recall: 16.5% -> 15.4% | Hurt |
| Include stop words in the subtitle vocabulary (ablation) | Video-to-text Transformer · BSL-1K testloc | recall: 16.5 -> 13.9 | Hurt |
| Use original subtitle words rather than stemming (ablation) | Video-to-text Transformer · BSL-1K testloc | recall: 16.5 -> 16.0 | No clear effect |
Closest papers by meaning
- BSL-1K: Scaling up co-articulated sign language recognition using mouthing cues (ECCV 2020 · similarity 0.94)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.91)
- Lost in Translation, Found in Context: Sign Language Translation with Contextual Cues (CVPR 2025 · similarity 0.91)
- TCNet: Continuous Sign Language Recognition from Trajectories and Correlated Regions (AAAI 2024 · similarity 0.89)
- Bridging Sign and Spoken Languages: Pseudo Gloss Generation for Sign Language Translation (NeurIPS 2025 · similarity 0.89)
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
